عنوان مقاله
معکوس CF: یک الگوریتم سریع فیلتر کردن مشترک با استفاده از یک نمودار K نزدیکترین همسایه
فهرست مطالب
مقدمه
کارهای مرتبط
فیلتر کردن مشترک سریع
آزمایش
نتیجه گیری
بخشی از مقاله
فیلتر کردن مشترک سریع
روش ما شامل دو مرحله اصلی است :
اول، ما حدود یک نزدیکترین گراف همسایهK^′ (نمودار K^′-NN) را به عنوان یک گام پیش پردازش می سازیم. دوم، ما K همسایگی آیتم های رتبه بندی نشده را بر اساس نمودار k0-NN پیدا می کنیم . سپس آیتم ها را به کاربران با استفاده از همسایگان K و نسخه اصلاح شده ما از همسایگی کسینوس غیر نرمالیزه شده ، توصیه می کنیم.
کلمات کلیدی:
Reversed CF: A fast collaborative filtering algorithm using a k-nearest neighbor graph Youngki Park a,⇑ , Sungchan Park a , Woosung Jung b , Sang-goo Lee a a Room 418, Building 138, Seoul National University, Sillim-9-dong, Gwanak-gu, Seoul, Republic of Korea b Room 206B, Building E8-10, Chungbuk National University, 12 Gaesin-dong, Heungdeok-gu, Cheongju-si, Chungcheongbuk-do, Republic of Korea article info Article history: Available online 10 January 2015 Keywords: Reversed CF Collaborative filtering k-Nearest neighbor graph Greedy filtering abstract User-based and item-based collaborative filtering (CF) methods are two of the most widely used techniques in recommender systems. While these algorithms are widely used in both industry and academia owing to their simplicity and acceptable level of accuracy, they require a considerable amount of time in finding top-k similar neighbors (items or users) to predict user preferences of unrated items. In this paper, we present Reversed CF (RCF), a rapid CF algorithm which utilizes a k-nearest neighbor (k-NN) graph. One main idea of this approach is to reverse the process of finding k neighbors; instead of finding k similar neighbors of unrated items, RCF finds the k-nearest neighbors of rated items. Not only does this algorithm perform fewer predictions while filtering out inaccurate results, but it also enables the use of fast k-NN graph construction algorithms. The experimental results show that our approach outperforms traditional user-based/item-based CF algorithms in terms of both preprocessing time and query processing time without sacrificing the level of accuracy. 2015 Elsevier Ltd. All rights reserved. 1. Introduction User-based and item-based collaborative filtering (CF) methods are two of the most widely used techniques in recommender systems. When a user requests a recommendation, the user-based CF algorithm, introduced by Herlocker, Konstan, Borchers, and Riedl (1999), predicts the users preferences for all of the unrated items based on similar users’ preferences for those items. In a similar way, the item-based CF algorithm presented by Sarwar, Karypis, Konstan, and Riedl (2001) predicts the preferences of the user for all unrated items based on the user’s preference levels for similar items. Cremonesi, Koren, and Turrin (2010) and Lee, Song, Kahng, Lee, and Lee (2011) state that CF algorithms produce movie recommendations of a higher quality compared to baseline algorithms, which only recommend the most popular movies or highly rated movies. Although there have been proposed more efficient algorithms, such as those that use singular vector decomposition presented by Cremonesi et al. (2010) or a random walk proposed by Lee et al. (2011) and Lee, Park, Kahng, and Lee (2013), CF algorithms are still widely used in both industry and academia owing to their simplicity and acceptable levels of accuracy. For example, the Amazon and YouTube recommender systems, introduced by Linden, Smith, and York (2003) and Davidson et al. (2010) respectively, utilize CF-based algorithms. Additionally, many modified versions of CF algorithms such as the work of Lee, Park, Kahng, Lee, and Lee (2010b) and Park, Lee, and Lee (2011) are being proposed for the purpose of building context-aware recommender systems.
